[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CA+X1aCSMb9CTpoHWv=UiMg-mHkB2ssmKRACtbDofWG=QSxXOdg@mail.gmail.com>
Date: Wed, 31 Aug 2011 13:12:02 +0400
From: Maxim Patlasov <maxim.patlasov@...il.com>
To: Shaohua Li <shli@...nel.org>
Cc: axboe@...nel.dk, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 1/1] CFQ: fix handling 'deep' cfqq
Shaohua,
Sorry for not getting back to you for long. See please inline comments below.
> but the cfq_disk_looks_slow isn't updated if the queue doesn't have 4 requests
> or doesn't dispatch > 4 requests, so you always have CFQD_DISK_LOOKS_FAST()
> return true if the first slice gets it to true. And if the queue does
> dispatch > 4 requests in one jiffy, only cfq_disk_looks_fast is updated,
> CFQD_DISK_LOOKS_FAST returns true too. I don't understand when
> CFQD_DISK_LOOKS_FAST can be false.
The patch essentially gathers events when cfqq experiences deep queue
early and dispatches 4 requests in one dispatch round. As soon as such
an event detected, we're drop inside outer 'if' clause:
> + if (cfq_cfqq_deep_early(cfqq) && cfqq->n_dispatched >= CFQQ_DEEP_THR) {
> + if (cfqq->first_dispatch == jiffies)
> + cfqd->cfq_disk_looks_fast++;
> + else
> + cfqd->cfq_disk_looks_slow++;
> +
> + cfqq->first_dispatch = 0;
> + cfqq->n_dispatched = 0;
> + cfq_clear_cfqq_deep_early(cfqq);
> + cfqd->cfq_disk_last_updated = jiffies;
> + }
and either increment disk_looks_fast or disk_looks_slow.
If the queue doesn't have 4 requests, cfqq is not 'deep' - no events
to gather. Neither disk_looks_fast nor disk_looks_slow is updated.
If the queue doesn't dispatch 4 requests in a raw, the event will be
discarded, and again, neither disk_looks_fast nor disk_looks_slow is
updated.
If the queue does dispatch > 4 requests in one jiffy, only
cfq_disk_looks_fast is updated - that's right. But if the queue
dispatches first 4 requests in *more* than one jiffy,
cfq_disk_looks_slow is updated.
BTW, note that CFQD_DISK_LOOKS_FAST returns true only if
disk_looks_fast greater than disk_looks_slow. So, in case of rare
accidental events, disk_looks_fast/slow counters will expire soon and
CFQD_DISK_LOOKS_FAST will return false ("0 > 0").
Thanks,
Maxim
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists